Merge f228ab54b9 into 7a445ee88d

While working on https://github.com/guardian/gha-scala-library-release-workflow I noticed that no matter how many times I ran the workflow, `actions/setup-java` would always report `sbt cache is not found`, even if there had been no substantial change in the project - simply that `version.sbt` (the file used by https://github.com/sbt/sbt-release) had the version number in it incremented (as in b2152325ba).

This meant that turning on `cache: sbt` would actually slow the workflow considerably, as it would never benefit from the cache being present, and would always have to save it, which could take 2-3 minutes - even though it can't take advantage of the data it's saving.

As such, it would be great to exclude `version.sbt` files from the cache hash key.

Background: `cache: sbt` was orginally introduced with https://github.com/actions/setup-java/pull/302

docs/advanced-usage.md

@@ -463,7 +463,7 @@ The result is a Toolchain with entries for JDKs 8, 11 and 15. You can even combi
     architecture: x64
 ```
 
-This will generate a Toolchains entry with the following values: `version: 1.6`, `vendor: jkdfile`, `id: Oracle_1.6`.
+This will generate a Toolchains entry with the following values: `version: 1.6`, `vendor: jdkfile`, `id: Oracle_1.6`.
 
 ### Modifying The Toolchain Vendor For JDKs
 Each JDK provider will receive a default `vendor` using the `distribution` input value but this can be overridden with the `mvn-toolchain-vendor` parameter as follows.

src/cache.ts

@@ -58,7 +58,8 @@ const supportedPackageManager: PackageManager[] = [
       '**/*.sbt',
       '**/project/build.properties',
       '**/project/**.scala',
-      '**/project/**.sbt'
+      '**/project/**.sbt',
+      '!**/version.sbt' // releasing a new version of a library project shouldn't invalidate the entire sbt cache
     ]
   }
 ];